This paper tries to redo the model from the Bogan’s paper (Bogan,2014) with a more reasonable dataset. Although the data I used is insufficient large, it is convinced to use since the dataset from Bogan’s paper is nearly the same size. The time, however, is 2007 -2009 versus 2011 – 2015, which make more sense considering the financial crisis in 2008. By comparing the result of two different models, I have similar findings regarding having elder person in the household and also the number of kids on holding safe assets and risky assets.
As we could expect, there are many things could influence the decision of household about the next generation’s education. Despite the main factor would be the difference of their class, the education level or their parents, or the ethnicity/race, there are still a lot of interest factors which could influence the assets’ allocation of the household, especially the educational savings for the next generations. Although it has been researched by Vicki L. Bogan in “Household Asset Allocation, Offspring Education, and the Sandwich Generation”, household with elder dependent will tend to reduce the risk assets which indicate the reduction of next generations’ education saving, the data set being used is 2007 – 2009, which may cause some effects due to the financial crisis in 2008.
Intuitively, people would tend to invest more in the safe asset while the expectation of the future is not well. In this case, having elder person in the household will decrease decreases the probability of risky asset holding by 0.129 (over twice as much as the household head having poor health) and decreases college savings account holding by 0.012 (twice as much as the household head having poor health) (Bogan, 2014). The reason I believe the result should not be representative is because the effect of the 2008 financial crisis seems not be considered in the paper. In the contrary, I will try to use the later data set to redo the process. What I expect is during year 2011 – 2015, there will be more representative result to explain what a household with elder person would influence the assets allocation, especially for next generations’ education savings.
What I find is quite interesting. The result from those two different databases is very similar which indicates that how people decide the assets allocation of their family is not influence by the financial crisis, or we can say it influenced, but not in a very considerable dimension. From the paper “The Impact of Skewness and Fat Tails on the Asset Allocation Decision” by James X. Xiong and Thomas M. Idzorek, there should be a different allocation comparing with other time (James and Thomas, 2011), which leads to the idea that why there are two contrary results. The possible reason is the different to define the variables, the different model used and different database.
The data I used is the 2016 SCF (Survey of Consumer Finances) which is the most recent survey conducted by the Federal Reserve System. Due to the large amount of questions containing in the survey, the sample size is just 31,240, not many people are willing to answer all the questions. The original data contains more than 5,000 variables, and most of the them is categorized by “yes” or “no”).
The average age of this sample is 52. Only 40 percent of them are categorized as married right now, others may be divorced, widowed or others. Approximately 69 percent of the them have own at least one home. 93 percent of the household holds safe assets, while 30 percent of the household holds risky assets and nearly 3 percent of the household hold education saving account. In the situation I defined in this paper, dependent elder who does not have savings for the retirement or any emergencies and also over 65 years old is 20 percent of the household. The reason I defined dependent elder in this situation is because that over 65, people are more likely to retire and live without any wages. So, people without having any savings for emergencies and retirement would unlikely to solve the problem by his own, which indicates they need depend on someone whom more likely be their children.
The variables I used for the final dataset are “id”, “age”, “kids”, “cgrad”, “married”, “black”, “hisp”, “employed”, “loginc”, “homeown”, “elderdum”, “Eexpcos”, “healthins”, “nothealthy”, “safe”, “risky” and “eduplan”. Among those variables, “cgrad”, “married”, “black”, “hisp”, “employed”, “homeown”, “elderdum”, “Eexpcos”, “healthins”, “nothealthy”, “safe”, “risky” and “eduplan” are dummy variables).
“id” is ascending on the year which is 2011 to 2015 and the household who took the survey. Normally, the last number will represent the people and year according to this manner. “age” represents the age of household head, and to avoid any problems, like people may cross section to be count as different group, we assume the age is same for the last year. In other words, the age appears in the data represent the age of the head of household in 2015. “kids” represents to the number of children, including natural children, step children and foster children. To define “cgrad”, we first collect the information from all heads’ education levels, then categorized those levels as four type. (1: no high school diploma, 2: high school diploma, 3: some college or Assoc degree, 4: Bachelor’s degree or higher). Then we took all 3’s and 4’s as 1 in the “cgrad”, 0 otherwise. “married” represents the marital status of the head of household and we count the first two types as married. (1. Married; 2. Living with a partner; 3. Separated; 4. Divorced; 5. Widowed; 6. Never married).
“black” and “hisp” represent black or African-American non-Hispanic and Hispanic or Latino respectively. “employed” represents those who work for someone else not self-employed, partnership with others, retired, disabled or under 65 and not working. “loginc” represents the log of the household income in previous calendar year. “homeown” represents whether the household has the house or mobile home. “elderdum” represents those who are over 65 years old and don’t have savings for the retirement or any emergencies, such like illness, medical or dental expenses. “Eexpcos” represents that there is a considerable amount savings for the expect cost of children’s’ future or college education. “healthins” represents those who have health insurance or have saving for future investment. “nothealthy” represents those who have poor healthy. “safe” represent all types of transactions accounts (including checking accounts, saving accounts, money market accounts, prepaid accounts and call accounts.), certificates of deposit, bonds (including tax-exempt bonds, mortgage-backed bonds and US government and government agency bonds and bills.). “risky” represent mutual funds (including stock mutual funds, tax-free bond mutual funds, government bond mutual funds and other bond mutual funds.) and all kinds of stocks. “eduplan” represents that household has education saving account or 529 plans.
The model we use is similar with the one appears in the paper I found, but what I do is not only using probit model, but also using logitistic model. I try to explain the result better by comparing those two models and hopefully to have a different result as I expected. The three independent variables are number of kids, whether or not having a remarkable amount of kids’ education expense as dummy variable and the whether or not having elder person in the household as dummy variable. The three dependent variables are holding safe assets as dummy variable, holding risky assets as dummy variable and having education saving account as dummy variable.
The control variables are those been shown be influenced in household investment and college savings behavior : log of household income (Bertaut, 1998), total household size (Keister, 2003), a respondent married dummy variable (DeVaney & Chien, 2002), a respondent age variable (Yilmazer, 2008), a respondent college graduate dummy variable (Lee & Hanna, 1995; Lefebvre, 2004), a respondent employed dummy variable, a respondent managerial or professional occupation dummy variable (Bogan, 2008), a home owner dummy variable (Babiarz & Yilmazer, 2001; Lefebvre, 2004), a poor health dummy variable (Rosen & Wu, 2004; DeVaney & Chien, 2002), a has health insurance dummy variable (Bogan & Fertig, 2013), a year 2009 dummy variable, and race dummy variables (Bogan, 2013).
For probit model, the reason I use it is that all dependent variables are binary variables. If we use linear regression model, the result will show the model and data doesn’t fit. If we can transform dependent variables into continuous variables, then we take those transformed variables as probability between 0 and 1 as a cumulative normal distribution ?. Then from Y= ?(X? + ?), we can conclude ?-1(Y) = X? + ? or in this case,
?-1(Yit) = ?0 + ??ktXikt + ?it.
Where i refers to safe assets, risky assets and education saving account, k refers to those three dependent variables, and t refers different time periods, which in this case year. By running probit model in STATA, we find out that in this case, the data fits all three probit models very well (the models are statistically significant because the p-value is less than .000). here is the table that summarize all three tables.
As we can see from the summary table of probit model, the number of kids will decrease the probability of holding safe assets, while the number of kids will increase the probability of holding risky assets and education saving accounts. (all of them are significant at 0.01 level). Having a remarkable amount of kids’ education expense in the future will decrease the probability of holding safe assets, while the expect education expense will increase the probability of having educational saving accounts. (expect cost for future education expense to hold safe assets is significant at 0.1 level, while having educational saving accounts is significant at 0.01 level). Having elder person in the household will increase the probability of holding safe assets, while having elder person in household will decrease the probability of holding risky assets and having educational saving accounts. (all of them are significant at 0.01 level).
To have a better idea about the result, I then choose to use logistic model to enforce the result I have above. The reason I use logistic model rather than logit model is that logistic model is easier to interpret. Since the dependent variables are binary variable with values 0 and 1, if we let p = E(Y|X) then the logistic model would be log(p/(1-p)) = ?0 + ??ktXikt, where ?0 is the odds ratio of constant, ??ktXikt is the summation of the odds ratio times three different independent variables.
VARIABLES safe risky eduplan
kids -0.172*** 0.157*** 1.079***
(0.0422) (0.0370) (0.130)
Eexpcos -0.233*** 0.0399 0.491***
(0.0806) (0.0588) (0.117)
elderdum 0.373*** -0.553*** -0.973***
(0.0905) (0.0465) (0.160)
Constant -11.21*** -6.590*** -7.244***
(0.429) (0.186) (0.435)
Observations 30,983 30,983 30,983
Robust standard errors in parentheses
A professional writer will make a clear, mistake-free paper for you!Get help with your assignment
Please check your inbox
I'm Chatbot Amy :)
I can help you save hours on your homework. Let's start by finding a writer.Find Writer